A New Method of Selecting Pivot Features for Structural Correspondence Learning in Domain Adaptive Sentiment Analysis

In recent years,Structural Correspondence Learning (SCL) is becoming one of the most important techniques for domain adaptation in natural language processing.T-SCL method for sentiment classification selects high frequency features which don´t have enough ability to discriminate positive instances from negative instances.Therefore, FisherA&IG-SCL method,a new method for selecting pivot features, is proposed. This method makes pivot features selected by Criterion function and Information Gain more discriminative and descriptive. The experimental results show that proposed FisherA&IG-SCL method can produce much better performance.